Welded pipe production process based on weld joint nondestructive testing

Abstract

The invention relates to the technical field of welded pipe production processes, and provides a welded pipe production process based on weld joint nondestructive testing, and the welded pipe production process based on weld joint nondestructive testing comprises the following steps: S1, strip steel pretreatment; s2, curling and forming; s3, welding is conducted, specifically, the joints of the pipe blanks are welded together to form a welded pipe; s4, nondestructive detection of the welding seam: s41, image acquisition; s42, carrying out image preprocessing; s43, defect detection and result output; s5, weld defect repairing: repairing the weld defect according to the defect detection result information; s6, sizing and straightening: sizing and straightening the welded pipe; s7, finished product inspection is conducted, specifically, appearance, size and mechanical property inspection is conducted on the welded pipes, and qualified welded pipes are screened out; according to the welded pipe production process, high-precision detection and real-time repair of weld defects in the welded pipe production process can be achieved, and the product quality and the production efficiency are improved.

Description

Technical Field

[0001] This application relates to the technical field of welded pipe production processes, and more specifically, to a welded pipe production process based on non-destructive testing of welds. Background Art

[0002] Seamless steel pipes, also known as welded pipes, are a major category of steel pipes, including straight seam welded pipes and spiral welded pipes. These steel pipes are manufactured by bending steel plates or strips into a predetermined shape and then refined through a welding process.

[0003] During the production of welded pipes, due to various factors (such as improper welding processes, material defects, operational errors, etc.), defects such as cracks, pores, slag inclusions, and lack of fusion may occur inside or on the surface of the welds. These defects not only affect the mechanical properties of the welded pipes but may also cause safety accidents such as leakage and fracture during use, resulting in serious economic losses and even casualties. Therefore, it is of great significance to detect and handle these defects in a timely manner through non-destructive testing of welds to ensure the quality of welded pipes and engineering safety.

[0004] In the traditional welded pipe production process, non-destructive testing of welds mainly relies on methods such as manual visual inspection and radiographic testing, which have problems such as low efficiency, high cost, and easy missed detection, resulting in low production efficiency and low product quality in the traditional welded pipe production process. Summary of the Invention

[0005] In view of this, this application provides a welded pipe production process based on non-destructive testing of welds to solve the technical problems of low production efficiency and low product quality in the existing welded pipe production process.

[0006] This application provides a welded pipe production process based on non-destructive testing of welds, wherein the welded pipe production process based on non-destructive testing of welds includes the following steps: S1. Strip pre-treatment: Uncoil, level, and shear the strip to obtain the pre-treated strip; S2. Curling and forming: Curling the pre-treated strip into a pipe blank; S3. Welding: Welding the joints of the pipe blank together to form a welded pipe; S4. Non-destructive testing of welds: including the steps: s41. Image acquisition: Real-time acquisition of the weld defect images of the welded pipe, and the acquired weld defect images are color images; s42. Image pre-processing: including denoising, enhancement, and segmentation processing of the weld defect images acquired in step s41; s43. Defect detection and result output: Placing the pre-processed image into the weld detection model, and the weld detection model outputs the weld defect detection result information; S5, Weld Defect Repair: Repair the weld defects according to the defect detection result information; S6, Sizing and Straightening: Size and straighten the welded pipe; S7, Final Inspection: Inspect the appearance, dimensions, and mechanical properties of the welded pipe, and screen out the qualified welded pipes; S8, Packaging and Warehousing: Package and label the qualified welded pipes and store them in the warehouse.

[0007] Furthermore, the weld detection model is constructed according to the following steps: Collect multi-modal data of the welded pipe weld, where the multi-modal data includes laser images and ultrasonic detection data; Preprocess the collected multi-modal data, including: performing median filtering and mean filtering denoising on the laser images, and performing time-frequency transformation on the ultrasonic detection data; Extract weld features: including extracting laser image features using a convolutional neural network and extracting ultrasonic time-frequency features using a recurrent neural network; Multi-modal Feature Fusion: Weightedly fuse the laser image features and ultrasonic time-frequency features through an adaptive attention mechanism to generate fused features; Construct an adaptive spatio-temporal attention network: Divide the weld area into multiple nodes, construct an adjacency matrix based on the similarity of node features, and use the spatio-temporal attention mechanism to dynamically weight the node features to capture the spatio-temporal features of weld defects; Defect Detection and Location: Use a fully connected layer to classify the node features and output the defect category, and use a fully connected layer to regress the node features and output the defect location; Model Training and Optimization: Use a classification loss function to optimize the defect classification task, use a regression loss function to optimize the defect location task, and evaluate and optimize the model performance through the test set and validation set.

[0008] Furthermore, the collection of multi-modal data of the welded pipe weld includes: using a laser scanner to collect laser images of the weld, and using an ultrasonic probe to collect ultrasonic reflection signals inside the weld.

[0009] Furthermore, the extraction of laser image features using a convolutional neural network includes: Perform adaptive noise suppression and contrast enhancement preprocessing on the original laser images; Construct a deep convolutional neural network containing a dual-path feature extraction structure, where: the main path uses a sequence of residual modules constructed by depthwise separable convolutions, and the auxiliary path uses an atrous spatial pyramid pooling structure; Perform channel attention weighted fusion on the dual-path features through a feature calibration module; The network is trained using a multi-task joint loss function, including a feature reconstruction loss and a contrast metric loss.

[0010] Furthermore, the image preprocessing further includes the following steps: s321. Color conversion: Convert the weld defect image from a color image to a grayscale image; s322. Adaptive region of interest extraction: Use an adaptive threshold segmentation algorithm to extract the weld region and remove background noise; s323. Image enhancement: According to the grayscale distribution characteristics of the weld image, use an adaptive histogram equalization algorithm; s324. Noise suppression: Extract the morphological features of the weld image and use an adaptive median filtering algorithm to remove image noise while retaining the weld edge information; s325. Image sharpening: Extract the weld edge features and use an adaptive Laplacian operator to enhance the image edge and improve the clarity of the weld contour; s326. Image binarization: According to the grayscale distribution characteristics of the weld image, use an adaptive threshold segmentation algorithm to binarize the image.

[0011] Furthermore, the adaptive threshold segmentation algorithm includes the following steps: Calculate the image grayscale histogram: Statistically calculate the frequency of each gray level in the grayscale image; Determine the optimal segmentation threshold: Calculate the optimal segmentation threshold based on the maximum inter-class variance method, and divide the weld image into foreground and background, where the foreground is the weld region; Extract the weld region: According to the optimal segmentation threshold, perform binarization processing on the weld image to extract the weld region.

[0012] Furthermore, the adaptive histogram equalization algorithm includes the following steps: Calculate the local region histogram: Divide the weld image into multiple local regions and calculate the grayscale histogram of each local region; Calculate the local region mapping function: According to the grayscale histogram, calculate the grayscale mapping function of each local region; Image grayscale mapping: According to the grayscale mapping function, perform grayscale mapping on the image.

[0013] Furthermore, the extraction of the morphological features of the weld image includes the following steps: The edge detection algorithm is used to obtain the weld edge information, and the local gradient intensity and direction are calculated; according to the local gradient intensity and direction, the filtering window size of the adaptive median filtering algorithm is dynamically adjusted: in the area where the gradient intensity is the first intensity, the first filtering window is adopted, and in the area where the gradient intensity is the second intensity, the second filtering window is adopted, the first intensity is greater than the second intensity, and the size of the first filtering window is smaller than the size of the second filtering window.

[0014] Further, the sizing and straightening includes the following steps: Heating: The welded pipe is uniformly heated through a preheating device, and the heating temperature is 800°C - 1000°C; Sizing: The heated welded pipe is fed into a sizing device, and the sizing device adopts a multi-stage rolling method, and the reduction amount of each stage of rolling gradually decreases; Straightening: The sized welded pipe is fed into a straightening device, and the straightening device adopts a multi-point straightening technique, and the number of straightening rolls is greater than or equal to 5; Cooling: The straightened welded pipe is cooled.

[0015] Further, the reduction amounts of each stage of rolling in the multi-stage rolling method are 0.5mm, 0.3mm, 0.2mm, and 0.1mm respectively.

[0016] The beneficial effects of the welded pipe production process based on weld non-destructive testing provided by the present invention are: Compared with the prior art, in the welded pipe production process based on weld non-destructive testing provided by the present invention, through image acquisition, image preprocessing, defect detection and result output, and weld defect repair in weld non-destructive testing, compared with methods such as manual visual inspection and ray detection, high-precision detection and real-time repair of weld defects in the welded pipe production process can be realized, improving product quality and production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a flow chart of a welded pipe production process based on weld non-destructive testing according to an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] To facilitate the understanding of this application, the following will provide a more comprehensive description of this application with reference to the relevant drawings. One or at least three embodiments of this application are exemplarily shown in the drawings to make the understanding of the technical solutions disclosed in this application more accurate and thorough. However, it should be understood that this application can be implemented in many different forms and is not limited to the embodiments described below.

[0020] In the drawings of this application, the same or similar reference numerals correspond to the same or similar components; in the description of this application, it should be understood that if there are terms such as "upper", "lower", "left", "right", etc. indicating the orientation or positional relationship, they are based on the orientation or positional relationship shown in the drawings. This is only for the convenience of describing this application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, the terms describing the positional relationship in the drawings are only for illustrative purposes and cannot be understood as a limitation of this application. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0021] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of this application, the descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the quantity of the technical features indicated. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In addition, if the "and / or" or "and / or" appears throughout the text, its meaning includes three parallel scenarios. Taking "A and / or B" as an example, it includes the A scenario, or the B scenario, or the scenario where both A and B are satisfied simultaneously.

[0022] In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of those of ordinary skill in the art to implement. When the combination of technical solutions results in contradictions or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0023] See Figure 1 , a welded pipe production process based on weld non-destructive testing is provided. Among them, the welded pipe production process based on weld non-destructive testing includes the following steps: S1. Strip pre-treatment: Unroll, flatten, and shear the strip to obtain the pre-treated strip. A specific embodiment can be: After unrolling the strip, eliminate the curling stress through a multi-roll flattener (such as a nine-roll flattener), and control the flattening accuracy within ±0.2 mm / m; use a hydraulic guillotine shear for shearing, with the perpendicularity error of the cut ≤ 0.1 mm. For Q235B strip with a thickness of 6 mm, set the pressure of the flattening roll to 150 MPa and the shearing length to 1200 mm. This step can eliminate the internal stress of the material, ensure the subsequent curling and forming accuracy, and reduce the risk of welding deformation; S2. Coiling and forming: Coil the pretreated strip steel into a tube blank. Specifically, it can be coiled into a cylindrical or other-shaped tube blank through a strip steel coiling and forming machine. A specific embodiment can be: using a three-roll symmetric coiling machine, through the involute forming method, with the upper roll diameter of Φ300mm, the side roll diameter of Φ250mm, the coiling speed of 2m / min, coiling the strip steel with a width of 500mm into a tube blank with a diameter of Φ219mm, and the ovality ≤0.5%. The coiling and forming operation in this step has high forming accuracy, can avoid warping of the tube blank edge, and provides a uniform seam gap for subsequent welding; S3. Welding: Weld the seams of the tube blank together to form a welded pipe. Specifically, welding methods such as arc welding, high-frequency or low-frequency resistance welding can be used. Specifically, high-frequency resistance welding (HFW) can be adopted, with a frequency of 350kHz, a current of 8000A, and a welding speed of 15m / min; immediately spray water for cooling after welding, with a cooling rate of 50℃ / s. When welding X70 pipeline steel with a thickness of 8mm, the weld penetration reaches 80% of the plate thickness, and the width of the heat-affected zone ≤1.5mm. High-frequency welding can achieve rapid heating and reduce oxidation; spray water cooling inhibits grain coarsening and improves the toughness of the weld; S4. Nondestructive testing of weld seams: Includes the steps: s41. Image acquisition: High-resolution industrial cameras and linear array light sources can be used to collect the weld defect images of the welded pipe in real time. The collected weld defect images are color images; s42. Image preprocessing: Includes denoising, enhancement, segmentation processing, and other appropriate preprocessing of the weld defect images collected in step s41 to improve the image quality; s43. Defect detection and result output: Place the preprocessed image into the weld detection model, and the weld detection model outputs the weld defect detection result information, which is convenient for real-time output of the defect type (porosity, crack, etc.) and coordinates, providing accurate input for repair; S5. Weld defect repair: According to the defect detection result information, specifically, the known corresponding repair program can be automatically called to control the laser welding robot or plasma welding robot to repair the weld defect. Specifically, a laser welding robot (IPG YLS-6000) with a power of 3kW, a spot diameter of 0.2mm, and a scanning speed of 5m / min can be used; the current of the plasma welding torch is 150A, and the argon gas flow rate is 15L / min. For an unfused defect with a depth of 0.5mm, the weld penetration reaches 1.2mm after laser repair, and the hardness is restored to 95% of the base metal. This non-contact repair avoids secondary thermal damage, and the microstructure of the repair area is uniform; S6. Sizing and straightening: Sizing and straightening the welded pipe to achieve the specified dimensional and shape accuracy. A specific embodiment can be: adopting multi-stage roll pressing. The sizing machine consists of 4 three-roll stands, with a roll gap tolerance of ±0.05 mm; the straightening machine is equipped with 7 rolls, and the pressure is adjusted in stages (20 - 50 kN). After sizing, the outer diameter tolerance of the Φ219×8 mm welded pipe is ±0.08 mm, and the straightness is 0.3 mm / m. The advantage is that multi-stage progressive deformation avoids material stress concentration and improves dimensional stability. S7. Final product inspection: Inspecting the appearance, dimensions, and mechanical properties of the welded pipe, and screening out qualified welded pipes. S8. Packaging and warehousing: Packaging and labeling the qualified welded pipes and storing them in the warehouse.

[0024] In the welded pipe production process provided by the present invention based on weld non-destructive testing, through image acquisition, image preprocessing, defect detection, and result output in weld non-destructive testing, as well as subsequent weld defect repair, compared with methods such as manual visual inspection and ray detection, high-precision detection and real-time repair of weld defects in the welded pipe production process can be achieved, improving product quality and production efficiency.

[0025] According to an embodiment of the present application, the weld detection model is constructed according to the following steps: Collect multi-modal data of the welded pipe weld, where the multi-modal data includes laser images and ultrasonic detection data; Preprocess the collected multi-modal data, including: performing median filtering and mean filtering denoising on the laser images and performing time-frequency transformation on the ultrasonic detection data. Specifically, in operation, laser image processing includes: median filtering (window 3×3) to remove salt-and-pepper noise, mean filtering (window 5×5) to eliminate Gaussian noise, and the signal-to-noise ratio is increased to 35 dB. Ultrasonic time-frequency transformation includes: performing short-time Fourier transform (STFT) on the reflected signal, with a window length of 256 points and an overlap rate of 75%, and extracting features in the 0.5 - 15 MHz frequency band; Extract weld features: including using a CNN (Convolutional Neural Network) to extract laser image features and using a recurrent neural network to extract ultrasonic time-frequency features; Multi-modal feature fusion: Weightedly fuse the laser image features and ultrasonic time-frequency features through an adaptive attention mechanism to generate fused features; Constructing an Adaptive Spatiotemporal Attention Network: The weld area is divided into multiple nodes, and an adjacency matrix is constructed based on the similarity of node features. The spatiotemporal attention mechanism is used to dynamically weight the node features to capture the spatiotemporal features of weld defects. Specifically, the adaptive spatiotemporal attention network can specifically include three aspects: node division, adjacency matrix construction, and spatiotemporal attention mechanism. Node division: The weld image is divided into grid nodes of 32×32 pixels, and each node extracts a 256-dimensional feature vector. Adjacency matrix construction: The node correlation is calculated based on cosine similarity, and the threshold is set to 0.7 to construct a sparse adjacency matrix. Spatiotemporal attention mechanism: The time dimension focuses on the evolution trend of defects in continuous production batches, and the space dimension focuses on the differences between the defect area and the surrounding tissues; Defect Detection and Location: Use a fully connected layer to classify the node features and output the defect category, and use a fully connected layer to regress the node features and output the defect location; Model Training and Optimization: Use a classification loss function (cross-entropy loss) to optimize the defect classification task, use a regression loss function (smooth L1 loss) to optimize the defect location task, and evaluate and optimize the model performance through the test set and validation set.

[0026] In this embodiment, through multi-modal data complementarity (laser detection for surface defects and ultrasonic detection for internal defects), the spatiotemporal attention network dynamically focuses on key areas, and the detection accuracy is improved by more than 30% compared with traditional methods.

[0027] According to a specific embodiment of the present application, the multi-modal data for collecting the weld of the welded pipe includes: using a laser scanner to collect the laser image of the weld, and using an ultrasonic probe to collect the ultrasonic reflection signal inside the weld. Specifically, multi-sensor collaborative work can be used to achieve full-coverage detection of internal and external defects of the weld.

[0028] According to an embodiment of the present application, the use of a convolutional neural network to extract laser image features includes: Perform preprocessing of adaptive noise suppression and contrast enhancement on the original laser image; Construct a deep convolutional neural network containing a dual-path feature extraction structure, where: the main path adopts a sequence of residual modules constructed by depthwise separable convolution, and the auxiliary path adopts an atrous spatial pyramid pooling structure. Specifically, the main path consists of 8 depthwise separable residual modules, each module containing a 3×3 DSC layer + a 1×1 convolution, and the computational complexity is reduced to 1 / 9 of the standard convolution. In the auxiliary path, the atrous spatial pyramid contains parallel convolutions with dilation rates of 2, 4, and 6, and the receptive field is extended to 112×112 pixels; Perform channel attention weighted fusion on the dual-path features through a feature calibration module. Specifically, the feature calibration module adopts the SENet attention mechanism to perform channel weighting on the dual-path features; The network is trained using a multi-task joint loss function, which includes a feature reconstruction loss and a contrast metric loss.

[0029] According to an embodiment of the present application, the image preprocessing further includes the following steps: s321. Color conversion: Convert the weld defect image from a color image to a grayscale image; s322. Adaptive region of interest extraction: Use an adaptive threshold segmentation algorithm to extract the weld region and remove background noise; s323. Image enhancement: According to the grayscale distribution characteristics of the weld image, use an adaptive histogram equalization algorithm, which can enhance the contrast of the weld image to highlight weld details; s324. Noise suppression: Extract the morphological features of the weld image and use an adaptive median filtering algorithm to remove image noise while retaining weld edge information (retaining weld edge information while removing noise to improve the accuracy of subsequent defect detection); s325. Image sharpening: Extract the weld edge features and use an adaptive Laplacian operator to enhance the image edges and improve the clarity of the weld contour; s326. Image binarization: According to the grayscale distribution characteristics of the weld image, use an adaptive threshold segmentation algorithm to binarize the image, which is convenient for subsequent weld feature extraction and recognition.

[0030] In addition, the adaptive threshold segmentation algorithm includes the following steps: Calculate the image grayscale histogram: Count the frequency of each gray level in the grayscale image; Determine the optimal segmentation threshold: Calculate the optimal segmentation threshold based on the maximum inter-class variance method, divide the weld image into foreground and background, where the foreground is the weld region, traverse the gray levels 0 - 255, and calculate the inter-class variance σ =ω0ω1(μ0 - μ1) , select the threshold corresponding to the maximum σ For an image with a gray mean of 120, the optimal segmentation threshold is 145, which successfully separates the weld from the oxide layer background. The adaptive threshold overcomes the influence of uneven illumination and can improve the segmentation accuracy; Extract the weld region: According to the optimal segmentation threshold, perform binarization processing on the weld image to extract the weld region.

[0031] According to a specific embodiment of the present application, the adaptive histogram equalization algorithm includes the following steps: Calculate the local region histogram: Divide the weld image into multiple local regions, calculate the grayscale histogram of each local region. A specific embodiment of the local region division can be: The image is divided into blocks of 32×32 pixels, and the boundary regions are filled with mirror images to avoid edge effects; Calculate the local area mapping function: Calculate the gray mapping function of each local area according to the gray histogram; Image gray mapping: Perform gray mapping on the image according to the gray mapping function to enhance the image contrast.

[0032] According to a specific embodiment of the present application, the extraction of the morphological features of the weld image includes the following steps: Use an edge detection algorithm to obtain the weld edge information, and calculate the local gradient intensity and direction; according to the local gradient intensity and direction, dynamically adjust the filter window size of the adaptive median filtering algorithm: in the area where the gradient intensity is the first intensity, use the first filter window, and in the area where the gradient intensity is the second intensity, use the second filter window. The first intensity is greater than the second intensity, and the size of the first filter window is smaller than the size of the second filter window. This is equivalent to using a smaller filter window in the area with a larger gradient intensity to retain the weld edge details; in the area with a smaller gradient intensity, use a larger filter window to enhance the noise suppression effect. In this embodiment, the gradient adaptive filtering can retain the micron-level defect edges while removing noise, and can reduce the false detection rate.

[0033] According to a specific embodiment of the present application, the sizing and straightening includes the following steps: Heating: Uniformly heat the welded pipe through a preheating device, and the heating temperature is 800°C - 1000°C, which can ensure the overall temperature of the pipe is uniform; Sizing: Feed the heated welded pipe into the sizing device. The sizing device adopts a multi-stage roll pressing method, and the reduction amount of each stage of roll pressing gradually decreases to ensure that the outer diameter accuracy of the pipe is controlled within ±0.1 mm. The multi-stage roll pressing makes the plastic flow of the material uniform; Straightening: Feed the sized welded pipe into the straightening device. The straightening device adopts a multi-point straightening technology. The number of straightening rolls is greater than or equal to 5, and the pressure of each straightening roll can be independently adjusted to ensure that the straightness error of the pipe is less than 0.5 mm / m; Cooling: Cool the straightened welded pipe. The cooling medium is water mist or air, and the cooling rate is controlled at 10°C / s - 20°C / s to ensure the uniformity of the internal structure of the welded pipe material.

[0034] According to a preferred embodiment of the present application, the reduction amounts of each stage of roll pressing in the multi-stage roll pressing method are 0.5 mm, 0.3 mm, 0.2 mm, and 0.1 mm respectively. In this embodiment, the progressive reduction is used to avoid the folding defect of the welded pipe wall and reduce the surface roughness.

[0035] It should be noted that the above embodiments only represent the preferred embodiments of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation of the present application. It should be pointed out that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can be made, such as combining different features in each embodiment, etc. These should all fall within the protection scope of the present application.

Claims

1. A production process for welded pipes based on non-destructive testing of welds, characterized in that, The steel pipe production process based on weld non-destructive testing includes the following steps: S1. Strip pre-treatment: Uncoil, level, and shear the strip to obtain the pre-treated strip. S2. Curling and forming: Curl the pre-treated strip into a tube blank. S3. Welding: Weld the seams of the tube blank together to form a steel pipe. S4. Weld non-destructive testing: It includes the steps of: s41. Image acquisition: Real-time acquire the weld defect images of the steel pipe, and the acquired weld defect images are color images. s42. Image pre-processing: It includes denoising, enhancement, and segmentation processing of the weld defect images acquired in step s41. s43. Defect detection and result output: Place the pre-processed image into the weld detection model, and the weld detection model outputs the weld defect detection result information. S5. Weld defect repair: Repair the weld defects according to the defect detection result information. S6. Sizing and straightening: Size and straighten the steel pipe. S7. Final inspection: Inspect the appearance, dimensions, and mechanical properties of the steel pipe, and screen out the qualified steel pipes. S8. Packaging and warehousing: Package and label the qualified steel pipes and store them in the warehouse.

2. The production process of welded pipes based on non-destructive testing of weld seams according to claim 1, characterized in that, The weld detection model is constructed according to the following steps: Collect multi-modal data of the steel pipe weld, and the multi-modal data includes laser images and ultrasonic detection data. Pre-process the collected multi-modal data, including: performing median filtering and mean filtering denoising processing on the laser images, and performing time-frequency transformation processing on the ultrasonic detection data. Extract weld features: It includes using a convolutional neural network to extract laser image features and using a recurrent neural network to extract ultrasonic time-frequency features. Multi-modal feature fusion: Dynamically weight and fuse the laser image features and ultrasonic time-frequency features through an adaptive attention mechanism to generate fusion features. Construct an adaptive spatio-temporal attention network: Divide the weld area into multiple nodes, construct an adjacency matrix based on the similarity of node features, and use the spatio-temporal attention mechanism to dynamically weight the node features to capture the spatio-temporal features of weld defects. Defect detection and localization: Use a fully connected layer to classify the node features and output the defect category, and use a fully connected layer to regress the node features and output the defect location. Model training and optimization: Use a classification loss function to optimize the defect classification task, use a regression loss function to optimize the defect localization task, and evaluate and optimize the model performance through the test set and validation set.

3. The production process of welded pipes based on non-destructive testing of welds according to claim 2, characterized in that, The collection of multi-modal data of the steel pipe weld includes: using a laser scanner to collect the laser images of the weld, and using an ultrasonic probe to collect the ultrasonic reflection signals inside the weld.

4. The production process of welded pipes based on non-destructive testing of welds according to claim 2, characterized in that, The use of a convolutional neural network to extract laser image features includes: Perform adaptive noise suppression and contrast enhancement pre-processing on the original laser image. Construct a deep convolutional neural network containing a dual-path feature extraction structure, where: the main path adopts a sequence of residual modules constructed by depthwise separable convolutions, and the auxiliary path adopts an atrous spatial pyramid pooling structure. Perform channel attention weighted fusion on the dual-path features through a feature calibration module. The network is trained using a multi-task joint loss function, which includes a feature reconstruction loss and a contrast metric loss.

5. The production process of welded pipes based on non-destructive testing of welds according to claim 1, characterized in that, The image preprocessing further includes the following steps: s321. Color conversion: Convert the weld defect image from a color image to a grayscale image; s322. Adaptive region of interest extraction: Use an adaptive threshold segmentation algorithm to extract the weld region and remove background noise; s323. Image enhancement: According to the grayscale distribution characteristics of the weld image, use an adaptive histogram equalization algorithm; s324. Noise suppression: Extract the morphological features of the weld image and use an adaptive median filtering algorithm to remove image noise while retaining the weld edge information; s325. Image sharpening: Extract the weld edge features and use an adaptive Laplacian operator to enhance the image edge and improve the weld contour clarity; s326. Image binarization: According to the grayscale distribution characteristics of the weld image, use an adaptive threshold segmentation algorithm to binarize the image.

6. The production process of welded pipes based on non-destructive testing of weld seams according to claim 5, characterized in that, The adaptive threshold segmentation algorithm includes the following steps: Calculate the image grayscale histogram: Count the frequency of each gray level in the grayscale image; Determine the optimal segmentation threshold: Calculate the optimal segmentation threshold based on the maximum inter-class variance method, and divide the weld image into foreground and background, where the foreground is the weld region; Extract the weld region: According to the optimal segmentation threshold, perform binarization processing on the weld image to extract the weld region.

7. The production process of welded pipes based on non-destructive testing of welds according to claim 5, characterized in that, The adaptive histogram equalization algorithm includes the following steps: Calculate the local region histogram: Divide the weld image into multiple local regions and calculate the grayscale histogram of each local region; Calculate the local region mapping function: According to the grayscale histogram, calculate the grayscale mapping function of each local region; Image grayscale mapping: According to the grayscale mapping function, perform grayscale mapping on the image.

8. The production process of welded pipes based on non-destructive testing of welds according to claim 5, characterized in that, The extraction of the morphological features of the weld image includes the following steps: Use an edge detection algorithm to obtain the weld edge information and calculate the local gradient intensity and direction; According to the local gradient intensity and direction, dynamically adjust the filtering window size of the adaptive median filtering algorithm: In the region with the first intensity of the gradient, use the first filtering window, and in the region with the second intensity of the gradient, use the second filtering window, where the first intensity is greater than the second intensity, and the size of the first filtering window is smaller than the size of the second filtering window.

9. The production process of welded pipes based on non-destructive testing of welds according to claim 1, characterized in that, The sizing and straightening includes the following steps: Heating: Uniformly heat the welded pipe through a preheating device, and the heating temperature is 800°C - 1000°C; Sizing: Feed the heated welded pipe into the sizing device, and the sizing device uses a multi-stage rolling method, and the reduction amount of each stage of rolling gradually decreases; Straightening: Feed the sized welded pipe into the straightening device, and the straightening device uses a multi-point straightening technology, and the number of straightening rolls is greater than or equal to 5; Cooling: Cool the straightened welded pipe.

10. The welded pipe production process based on weld nondestructive testing according to claim 9, characterized in that, In the multi-stage rolling method, the reduction amounts of each stage of rolling are 0.5mm, 0.3mm, 0.2mm, and 0.1mm respectively.

Images